Although microphones intended for conference tables aren’t suitable for inconspicuous hearing aids, they go a long way toward working out algorithms (*). This is a SoundTech CM-1000 USB mic:

It produces noise-canceled stereo output and a quick test shows impulse sounds produce reasonable left and right responses responses; I can’t vouch for the noise cancelling part.
A click to the right side:

And to the left:

The green trace (Channel 2) is obviously the Right channel, which corresponds to in1
on the Scope Sink block and out1
of the Audio Source in the GNU Radio data flow diagram:

There’s an irreconciliable clash between 0-index and 1-index numbering in there, but the microphone’s “Left” and “Right” channels appear in the proper places when you look at the mic from the conference room side of the label as shown in the top photo.
Figuring the speed of sound at 344 m/s, that 100 µs delay means the mic capsules sit 34 mm apart, which looks to be about right, as the flat part of the housing under the label spans 22 mm.
That’s a tad skimpy for things like beamforming and direction finding, so I actually bought a set with a separate CM-1000 mic that plugs into the USB mic:

The channel layout diagram explains what’s supposed to happen:

The additional mic changes the response, so that the USB unit becomes the Left channel and the analog mic provides the Right channel. I don’t know what happens to the “noise canceling” part of the story.
With the mics positioned 200 mm on center, a click to the right side:

The eyeballometrically precise 600 µs delay corresponds to 206 mm at 344 m/s, which might actually be close: they’re 200 mm on center, but the Right-channel mic is 10 mm smaller and the mic might be half that much further away from the other one. Not that that makes any difference.
(*) And, frankly, slapping a mic on the table won’t bother me much at all…